基于迁移学习的卷积神经网络通道剪枝

卷积神经网络在计算机视觉等多个领域应用广泛,然而其模型参数量众多、计算开销庞大,导致许多边缘设备无法满足其存储与计算资源要求。针对其边缘部署困难,提出使用迁移学习策略改进基于BN层缩放因子通道剪枝方法的稀疏化过程。本文对比不同层级迁移方案对稀疏化效果与通道剪枝选取容限的影响;并基于网络结构搜索观点设计实验,探究其精度保持极限与迭代结构的收敛性。实验结果表明,对比原模型,采用迁移学习的通道剪枝算法,在精度损失不超过0.10的前提下,参数量减少89.1%,模型存储大小压缩89.3%;对比原剪枝方法,将剪枝阈值从0.85提升到0.97,进一步减少参数42.6%。实验证明,引入迁移策略更易实现充分的稀疏化,提高通道剪枝阈值选取容限,实现更高压缩率;并在迭代剪枝的网络结构搜索过程中,提供更高效的搜索起点,利于快速迭代趋近至搜索空间的一个网络结构局部最优解。     

智能电网中电动汽车充电的自适应电价控制方法

针对电动汽车的充电提出一种电价控制策略。聚合管理者集中管理电动汽车的电池,并且考虑用电高峰时电网的电能供给有限,通过电价控制调整充电的需求量。采用自适应动态规划,通过在线网络训练,得到最优的电价策略。仿真结果表明,该自适应电价控制方法能够通过学习电动汽车的移动性和充电过程,从而调整实际充电需求量至期望水平,保证智能电网的稳定运行。     

Research and Application of Dynamic Equation for Full Frontal Impact

Full frontal impact theory needs researching and exploring to satisfy the primary safety design of occupant restraint system,avoiding the increasingly "engineering"trend in order to develop and design safety vehicle. After occupant restraint system is simulated by using linear elastic stiffness k,the occupant-vehicle frontal rigid barrier impact model is established. Dynamic equation of dummy chest coupling vehicle is built for full frontal impact based on ordinary vehicle deceleration by Hooke law,and the equation is solved by comparing coefficient and satisfying boundary qualifications. While relative vehicle characteristic parameters are kept unchanging,the actual vehicle deceleration is fitted to the simplified equivalent square wave( ESW),tipped equivalent square wave( TESW) and equivalent dual trapezoids wave( EDTW). Phase angle  and amplitude A of dynamic equations based on ESW,TESW and EDTW are calculated and deduced. The results show that: the dynamic equation of dummy chest coupling vehicle can be well utilized to instruct the primary safety design of full frontal impact for objective vehicle to satisfy chest deceleration demands and the equation based on TESW is best for this design.     

Finite difference method for dynamic response analysis of anchorage system简

Based on some assumptions, the dynamic analysis model of anchorage system is established. The dynamic governing equation is expressed as finite difference format and programmed by using MATLAB language. Compared with theoretical method, the finite difference method has been verified to be feasible by a case study. It is found that under seismic loading, the dynamic response of anchorage system is synchronously fluctuated with the seismic vibration. The change of displacement amplitude of material points is slight, and comparatively speaking, the displacement amplitude of the outside point is a little larger than that of the inside point, which shows amplification effect of surface. While the axial force amplitude transforms considerably from the inside to the outside. It increases first and reaches the peak value in the intersection between the anchoring section and free section, then decreases slowly in the free section. When considering damping effect of anchorage system, the finite difference method can reflect the time attenuation characteristic better, and the calculating result would be safer and more reasonable than the dynamic steady-state theoretical method. What is more, the finite difference method can be applied to the dynamic response analysis of harmonic and seismic random vibration for all kinds of anchor, and hence has a broad application prospect.     

注塑产品动态装配模型的集成知识模型研究

基于MBR(基于模型的推理)技术，建立了注塑模具动态装配结构模型；基于RBR(基于规则的推理)技术，应用面向对象的规则模型(OOR)，建立了动态装配规则的知识表示模型。最后，该文将两种知识表示模型集成，建立了完整的知识模型，为整个系统的后续工作奠定了基础。     

数控机床位置伺服在线调试与动态性能分析研究

介绍一种在数控机床位置伺服系统在线调试中应用系统动态性能分析方法的基本原理，给出了一种位置伺服系统动态响应曲线、动态性能分析及其主要技术指标的实例。     

工程设计演示文档动态性表述的探讨

在工程投标、方案论证、学术交流中 ,可利用多媒体演示软件制作各类文档。为了充分表达制作人员的思想 ,达到预期目的 ,可采用丰富多彩的动态性表现手法 ,使演示产生很好的效果 ,令听众和观众欣然接受。为此 ,对工程设计演示文档的动态性表述进行了探讨 ,以便能更充分地发挥多媒体演示软件的作用     

隔直电容在弧焊逆变器中抗偏磁能力的分析

本文采用状态平均法建立了隔直电容在知式焊接逆变器中抗偏磁的动态模型，并进行了动态分析，为焊接逆变器的设计提供了理论依据。     

高效动态存储再分配方案与实现

为提高大小和数量都不确定的数据动态存储的空间利用率和处理速度,针对常规系统动态存储分配存在的问题,提出一种高效的动态存储再分配方案。通过减少系统动态存储分配的执行次数,在有效提高存储空间利用率的同时,提高存储空间动态分配的速度,给出动态存储再分配方案的具体实现算法,用实际数据对其性能进行测试。测试结果表明,该方案能节省36％至75％的系统存储空间,分配速度提高了20至50倍。     

仿生波状前缘机翼动态失速控制的数值研究

该文提出了采用波状前缘机翼改善动态失速性能的有效控制方式。通过数值求解雷诺平均N-S方程,对比研究了波状前缘机翼和传统的光滑前缘机翼在动态失速时的力学性能和流动特征,并分析了前缘形状改进和可控拍动参数的影响,探讨有效的控制途径。研究结果表明,在深失速的动态失速控制中,波状前缘翼的复杂涡系能更稳定地附着在上翼面,其力学性能明显优于光滑前缘翼,尤其在大攻角范围内优势明显。波状前缘形状改变对升力能几乎没有影响,但转轴前移、增加拍幅和频率都能提高拍动波状前缘翼的升力。